HeartLock is a research/portfolio project exploring ECG waveform morphology as a biometric signal. It is not a production authentication system and must never be used to gate access to anything real. There is no liveness detection, no protection against replay of a recorded/synthesized ECG signal, no template encryption, and no defense against an adversary who can supply arbitrary input signals. A biometric feature this easy to bypass with a recording is not a security control.
Within that scope, the things this project does try to get right:
- No raw biometric data in the repository.
data/(all downloaded PhysioNet recordings) is gitignored from the first commit, before any data is ever pulled. - No secrets in the codebase. There are none needed for this project
(PhysioNet's open-access databases require no credentials), but
.envis gitignored in case that ever changes. - CLI input validation. Subject ids and session numbers supplied on the
command line are validated against a strict pattern before being used to
construct any file path or PhysioNet URL, to prevent path traversal (see
heartlock/data_loader.py::_validate_subject_id). - No
eval/exec/pickleon any input. Enrollment data is persisted withnumpy.savez/numpy.load(..., allow_pickle=False), neverpickle. - Pinned dependencies.
requirements.txtandpyproject.tomlpin exact versions to avoid unreviewed floating-version supply-chain surprises. - Static analysis.
banditis run over the codebase as part of the development workflow.
This is a personal portfolio project with no production deployment. If you spot a security issue anyway, please open an issue on the repository.